New work highlights how foundation-style multimodal models are transforming spatial multi-omic studies by integrating histopathology images with molecular maps to reveal cellular phenotypes in situ. Researchers demonstrated that pre-trained pathology models can be fine-tuned to enhance feature extraction and cross-modal inference across spatial transcriptomics and proteomics datasets. The approach reduces the need for bespoke models per dataset and improves transferability between cohorts, accelerating biomarker discovery and spatially informed target selection. The trend points to growing convergence between large-scale AI models and wet-lab spatial technologies in both academia and industry.